Right now, we get a ZeroDivisionError with SuccessiveHalving based algorithms because the fidelity parameter has a lower bound of 0 in a testing setup.
Moreover, perhaps we should explicitly check that it's positive all the time? This might belong in the search space instead though.
Right now, we get a
ZeroDivisionErrorwithSuccessiveHalvingbased algorithms because the fidelity parameter has a lower bound of0in a testing setup.Moreover, perhaps we should explicitly check that it's positive all the time? This might belong in the search space instead though.
https://github.com/automl/neps/blob/82b48903d0896db7d84ea2810f567a29d01688c3/neps/optimizers/multi_fidelity/successive_halving.py#L121-L123